Conclusions and recommendations

From the analyses performed thus far, it is clear that the ship data must be treated with great care in any satellite comparisons because of their high noise level. At the very least, the ship data should be screened to eliminate observations deviating unreasonably from climatology. In fact, a two pass screening is better, with the second pass screening out data deviating unreasonably from the first pass mapping attempt. Even with screening, the noise level is comparable with that of the satellite data and, particularly in spot comparisons, should not be used to evaluate the accuracy of the satellite data. Consideration also should be given to more sophisticated mapping procedures than simple binning averaging. Procedures that attempt local least squares fitting to individual observations would give better results because they would then be less sensitive to the spatial distribution of the data.